I wonder how hard it would be to turn entire yt channels into LLM fine tuning datasets.
Could toggle different topics
Thinking nostr:nprofile1qqsqqx9hacelkffcgd3ecchzjtlvwq9xn2fmprhrwnzmm2t3exee2eqpr9mhxue69uhhqun9d45h2mfwwpexjmtpdshxuet59uqsuamnwvaz7tmev9382tndv5hszxthwden5te0wfjkccte9eekummjwsh8xmmrd9skctcslkr8d's rumble channel could fuel a based AI, hell, all btc podcasts could be on the menu
Process is basically download, transcribe, chunk and have an LLM make QA pairs, then do a little LoRA fine tune
I already made a vector store of all Anthony chaffee's QA videos, took like a morning